Recent advances in genome sequencing, machine learning, and vaccine production technology has enabled cancer neoantigen vaccines, which allows cancer patients to be vaccinated against mutant proteins produced by their own tumors, stimulating the immune system to attack these cancer cells. Phase I clinical trials for Melanoma have shown remarkable results, both alone and with immune checkpoint inhibitors. I have developed a new method of predicting cancer neoantigens for cancer vaccines based on my understanding of tumor mutational signatures and cancer immunology. I have filed a provisional US patent, which University of California granted me full rights to. I am currently working to test if mutational signatures associated with specific types of cancer can retrospectively predict cancer neoantigens found in actual cancer patients and hope to publish these preliminary results soon. My proof-of-concept software implements this algorithm by scanning all human protein-coding genes for nucleotide sequence signature patterns for a specific COSMIC mutational signature. For each instance of a match, the peptide resulting from the amino acid change caused by the mutational signature mutation is analyzed by a machine-learning model for its MHC affinity and processing (using OpenVax MHCFlurry). The resulting whole-genome list of high affinity peptides would then be considered vaccine candidates for a person with that particular MHC allele and that particular mutational signature, although in clinical practice the actual calculated mutational signature would be derived from that person's actual mutational signature (from whole genome sequencing).
I am currently generating vaccine peptide datasets for all common MHC alleles in order to compare them to cancer neoantigens experimentally detected in cancer patients and hope to publish these results soon.
The core Python code can be found here.